China has just ranked second in intelligent computing capacity. The important figure is not the most striking

There’s a simple way to hype up artificial intelligence: just talk about models. And there is a more useful way to understand it: look at which countries have the capacity to train them, run them and bring them to millions of users without the system breaking along the way. In that second race, much less showy but much more revealing, China has just presented its numbers. The figure draws attention due to the ranking, yes, but the important thing is what it tells about the foundations of its deployment.

The figure. The information comes from Digital China Development Report 2025the document with which the National Data Administration summarizes China’s digital development over the past year. There it is maintained that the country reached 1.59 million PFLOPS in FP16 of intelligent computing capacity and that this volume would place it in second place in the world.

There is small print. The aforementioned report places China in second place in the world within a specific category: intelligent computing capacity. That is not the same as saying that the country is second in the entire AI race, where models, chips, talent, investment, adoption, regulation and many other variables come into play. What we are looking at is something more limited: the computing capacity prepared to power large-scale artificial intelligence loads.

Unity matters. FP16 stands for 16-bit floating point, a way of representing numbers with less precision than FP32 or FP64. It is widely used in artificial intelligence because it allows you to perform more operations and use less memory, a useful balance when we talk about training or running models. PFLOPS, for their part, serve to express how many floating point operations an infrastructure can perform every second.

It’s not just power. The report does not stop at 1.59 million PFLOPS FP16. It adds a very specific physical layer: more than 13.73 million standard racks in operation, 42 large intelligent computing clusters, described in the document as “ten thousand card” clusters, and a national testing and verification platform that already supervises 1,129 facilities. This network, according to the document, allows 110,000 PFLOPS to be coordinated for economic, scientific and government uses.

The distance with the US. China is placed second and the reference for first place is the United States. That is the reading that appears in Chinese state sources when they talk about a position “only behind the United States”, and also what external analyzes of high-end AI computing draw. The American advantage is not explained only by having more chips: also by data centers, large technology companies capable of financing enormous-scale infrastructures and a highly developed network.

The other half of the data is in the use. According to the report, China had 748 registered generative AI services at the end of 2025, of which 446 had been registered during that year. It also talks about 602 million users of generative AI, with a year-on-year growth of 141.7%. These are official figures and should be treated as such, but they help to understand why computing capacity matters: we are not dealing with an infrastructure designed only for laboratories, but for services that are already deployed on a large scale.

Images | Xataka with Nano Banana

In Xataka | Unitree is doing with robots what DJI did with drones: becoming inevitable